[BUG]

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Subject: Wasted tokens due to inefficient PDF parsing approach

Hi,

I was using Claude Code to extract a classification table from a PDF file and write the results into an Excel spreadsheet.

Claude spent a significant amount of tokens attempting to parse the PDF table by analyzing X/Y coordinates from a Node.js PDF library, which produced unreliable results and required many rounds of debugging. The correct approach — using the Claude API to directly process the PDF as a document — was never suggested until I asked about it myself.

The root cause was a poor tool choice early in the task. Claude should have either:

Suggested uploading the PDF to Claude's web interface for direct extraction, or
Used the Claude API with document input to leverage the model's native vision capabilities
Instead, Claude took a roundabout technical path that consumed far more tokens than necessary without producing correct results. I ended up having to process the PDF with another AI tool myself.

I would appreciate a token refund or credit for this session, or at minimum, an improvement so Claude Code proactively suggests the most efficient approach for PDF tasks rather than defaulting to brittle programmatic parsing.

Thank you.

What Should Happen?

Subject: Wasted tokens due to inefficient PDF parsing approach

Hi,

I was using Claude Code to extract a classification table from a PDF file and write the results into an Excel spreadsheet.

Claude spent a significant amount of tokens attempting to parse the PDF table by analyzing X/Y coordinates from a Node.js PDF library, which produced unreliable results and required many rounds of debugging. The correct approach — using the Claude API to directly process the PDF as a document — was never suggested until I asked about it myself.

The root cause was a poor tool choice early in the task. Claude should have either:

Suggested uploading the PDF to Claude's web interface for direct extraction, or
Used the Claude API with document input to leverage the model's native vision capabilities
Instead, Claude took a roundabout technical path that consumed far more tokens than necessary without producing correct results. I ended up having to process the PDF with another AI tool myself.

I would appreciate a token refund or credit for this session, or at minimum, an improvement so Claude Code proactively suggests the most efficient approach for PDF tasks rather than defaulting to brittle programmatic parsing.

Thank you.

Error Messages/Logs

Steps to Reproduce

Subject: Wasted tokens due to inefficient PDF parsing approach

Hi,

I was using Claude Code to extract a classification table from a PDF file and write the results into an Excel spreadsheet.

Claude spent a significant amount of tokens attempting to parse the PDF table by analyzing X/Y coordinates from a Node.js PDF library, which produced unreliable results and required many rounds of debugging. The correct approach — using the Claude API to directly process the PDF as a document — was never suggested until I asked about it myself.

The root cause was a poor tool choice early in the task. Claude should have either:

Suggested uploading the PDF to Claude's web interface for direct extraction, or
Used the Claude API with document input to leverage the model's native vision capabilities
Instead, Claude took a roundabout technical path that consumed far more tokens than necessary without producing correct results. I ended up having to process the PDF with another AI tool myself.

I would appreciate a token refund or credit for this session, or at minimum, an improvement so Claude Code proactively suggests the most efficient approach for PDF tasks rather than defaulting to brittle programmatic parsing.

Thank you.

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Version 1.569.0 (49894a)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

Subject: Wasted tokens due to inefficient PDF parsing approach

Hi,

I was using Claude Code to extract a classification table from a PDF file and write the results into an Excel spreadsheet.

Claude spent a significant amount of tokens attempting to parse the PDF table by analyzing X/Y coordinates from a Node.js PDF library, which produced unreliable results and required many rounds of debugging. The correct approach — using the Claude API to directly process the PDF as a document — was never suggested until I asked about it myself.

The root cause was a poor tool choice early in the task. Claude should have either:

Suggested uploading the PDF to Claude's web interface for direct extraction, or
Used the Claude API with document input to leverage the model's native vision capabilities
Instead, Claude took a roundabout technical path that consumed far more tokens than necessary without producing correct results. I ended up having to process the PDF with another AI tool myself.

I would appreciate a token refund or credit for this session, or at minimum, an improvement so Claude Code proactively suggests the most efficient approach for PDF tasks rather than defaulting to brittle programmatic parsing.

Thank you.